EM Sensor Machine Learning Recognition Power Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing EM sensors face increased power consumption and operational burden on main processors when identifying external electronic devices due to the need for continuous processing of electromagnetic signals, which can shorten battery life and increase processing load.
Innovation Solution
An EM sensor with a microcontroller unit and sensor memory that stores machine learning models, allowing for initial recognition and classification of electronic devices without main processor intervention, with feature values being transmitted only when initial models fail to recognize the device, thereby reducing processor load and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the main processor processes all EM signals for device recognition, then recognition accuracy is maintained, but power consumption increases and processor load increases
Solution Approach 1:
The patent segments the device recognition system into two parts: an EM sensor with integrated machine learning models for preliminary recognition, and a main processor for final confirmation. This segmentation allows the EM sensor to handle routine recognition tasks independently, reducing main processor involvement and power consumption while maintaining overall recognition accuracy through the hierarchical structure.
Solution Approach 2:
The EM sensor acts as an intermediary between the electromagnetic signal source and the main processor. It pre-processes EM signals using embedded machine learning models and only transmits feature values when recognition is uncertain or when specific conditions are met, thereby reducing the processing burden on the main processor while maintaining system-wide recognition accuracy.
2Reliability
If the main processor continuously processes EM signals, then device identification is reliable, but operational burden on processor increases
Solution Approach 1:
The EM sensor performs preliminary device recognition using its integrated machine learning models before involving the main processor. By pre-processing EM signals and identifying devices at the sensor level, the system maintains reliable device identification while significantly reducing the operational burden on the main processor, as it only needs to handle cases where preliminary recognition is uncertain.
Solution Approach 2:
The EM sensor is equipped with self-service capabilities through embedded machine learning models that enable it to independently perform device recognition tasks. This self-service functionality reduces dependency on the main processor, maintaining identification reliability while decreasing processor operational burden by handling routine recognition autonomously.
3Adaptability or versatility
If machine learning models are stored in main storage, then model availability is complete, but access time and power consumption increase
Solution Approach 1:
The patent implements a nested storage structure where machine learning models are stored in both the EM sensor's local memory and the main processor's storage. This nested arrangement allows the system to access models from the closer EM sensor memory for faster processing, while maintaining the complete model set in main storage for comprehensive adaptability, thereby reducing access time without sacrificing model availability.
Solution Approach 2:
The system applies local quality by storing frequently used or critical machine learning models in the EM sensor's local memory, while maintaining the complete model repository in main storage. This approach optimizes access time for common recognition tasks by keeping relevant models locally available, while preserving full model availability in main storage for less frequent or more complex recognition scenarios.
Data Source
AI summary
An electromagnetic (EM) sensor includes a front end module generating an EM signal using electromagnetic waves transmitted from an external source, a sensor memory storing a portion of a plurality of machine learning models used to recognize the EM signal, and a microcontroller unit for recognizing the external electronic device emitting the electromagnetic waves by inputting feature values extracted from the EM signal to the machine learning models. If the machine learning models stored in the sensor memory are not able to recognize the external device, the feature values may be transmitted to a main processor, and the main processor may compare the feature values to another set of machine learning models.